Exploring Adaptation in Dialog Systems

摘要: Numerous psycholinguistic studies have demonstrated that people adapt their language use in conversation to that of their conversational partners. Human conversational partners adapt to each others choice of words (particularly referring expressions), syntax, gesture and eye gaze. For example, if Susan uses the phrase my friend Mary to refer to her boss while talking with Sam, Sam will tend also to use your friend or Mary even if he knows that Mary is Susan’s boss. There is also some evidence that people adapt their language use in conversation with computer partners-spoken dialog systems. Spoken dialog systems allow users to access computer databases and accomplish tasks using speech (and sometimes multimodal input). The interaction with a dialog system is impaired by frequent speech recognition and understanding errors. The goal of much dialog systems research is to reduce the frequency of system errors and find strategies for minimizing their impact. In my thesis I investigate how adaptation is exhibited by users of dialog systems and how it may be used to reduce the frequency of speech recognition and understanding errors and recover from them more quickly. I explore variation in the users’ phrasing of concepts like dates and times and investigate whether users may be implicitly directed into using particular phrasings through interaction with the system. To do this, I use a combination of corpus analysis, dialog system engineering and experimental research. In my experimental research, I am planning user studies in which a user interacts with a dialog system. In one of my planned studies I will measure the effect of prompt …